The present invention relates to a digital watermark technique for use with digital data, particularly image data.
Recently, from the viewpoint of copyright protection for digital data, such as image data, etc., a digital watermark technique has been attracting significant attention. A digital watermark technique is defined as a technique which involves the embedding of specific information into digital data according to a predetermined rule so that the specific information cannot be extracted from the digital data without using at least the predetermined rule. According to this technique, information, for example, concerning a purchaser of image data, etc., is embedded in the image data according to a predetermined rule in such a manner that the information is not visible. Then, when the image data has been illegally copied, using the predetermined rule, the embedded information is extracted from the data that was illegally copied, to specify that the person has illegally copied the data.
FIG. 11 is a diagram showing the principle involved in the process of embedding/extracting information into/from image data using a conventional digital watermark technique.
With respect to the information embedding processing, as shown in FIG. 11, when a bit bi (1xe2x89xa6ixe2x89xa650) of bits b1-b50 constituting information to be embedded is equal to 1, the luminance of each pixel located at predetermined spaced positions of the image data is increased by U, and when the bit bi is equal to zero, the luminance of each such pixel is reduced by U. This processing is carried out for all of the bits b1-b50 while the embedding position is shifted, thereby embedding the digital watermark information in the image data.
On the other hand, the data value of the image data in which information was embedded, as stated above, may have been changed by conducting various image processings, including extension, reduction, compression, copying, etc. Therefore, with respect to extracting the embedded information, for the bit bi (1xe2x89xa6ixe2x89xa650) of bits b1-b50, the luminance sum S of the pixel data values located at the predetermined spaced positions of the image data and the luminance sum R (Reference value) of the pixel data values adjacent to the predetermined positions are detected. If Sxe2x88x92R greater than T (T is varied in accordance with a demanded error rate and it is set to satisfy T greater than U (the number of pixel data in which bi is embedded)), it is judged that bi is equal to 1. If Sxe2x88x92Rxe2x89xa6xe2x88x92T, it is judged that bi is equal to zero. If xe2x88x92T less than Sxe2x88x92R less than T, it is judged that no information is embedded in each of the pixel data located at the predetermined spaced positions. This processing is carried out for all of the bits b1 to b50 to extract the digital watermark information embedded in the image data.
The above-mentioned digital watermark technique is described in more detail in Kobayashi. et al. : Data Hiding Based on Neighbor Pixels Statistics, In Proc. of IPSJ 56th Annual Conference, IV-03, (1998).
In the process of extracting the digital watermark information based on the above-reference conventional digital watermark technique, the differential value between the sum S of the luminance values of pixels located at positions estimating that bit bi is embedded therein and the sum R of the luminance values of pixel data adjacent to the pixels respectively located at these positions is calculated. Then, it is judged that information (bit bi) 1 or 0 has been embedded, when the differential value is statistically significant, and it is judged that information (bit bi) has not been embedded, when the differential value is not statistically significant.
Due to the ruggedness of the luminance of image data, such as at contours of an image, a remarkable luminance difference may occur between this pixel data and adjacent pixel data. Therefore, such circumstances may cause Sxe2x88x92R greater than T or Sxe2x88x92R less than xe2x88x92T. In cases in which the ruggedness causes Sxe2x88x92R greater than T or Sxe2x88x92R less than xe2x88x92T, it may be judged that information has been embedded in the image data in cases where such information has not been embedded. Conversely the ruggedness may cause xe2x88x92T less than Sxe2x88x92R less than T in spite of the fact that information has embedded. In this case, it is incorrectly judged that information has not been embedded in the image data.
Although increasing the strength of U for embedding information and the criterion for extracting information (the threshold value T) can eliminate such misjudgments, this processing affects the quality of the image. The strength U is expressed by the equation U=xcex94xc3x97{square root over ( )}(V/N), in which V is the variance of Sxe2x88x92R, N is the number of pixel data of each bit, and xcex94 is the constant which depends on the detection accuracy. This equation concretely shows that the strength of U of the embedding information needs to be {square root over ( )}n times in order to keep the same detection accuracy, when the fluctuation (variance V) of Sxe2x88x92R produced by the ruggedness of the luminance of the image becomes n times. It also shows that the same detection accuracy can be maintained in spite of a small U, when the fluctuation (variance V) of Sxe2x88x92R becomes small.
The present invention is directed to the above-described circumstances, and an object of the present invention is to provide a digital watermark technique that can detect and extract embedded information precisely without increasing the strength of the embedding information.
Another object of the present invention is to provide a information embedding technique that can detect and extract embedded information precisely, while reducing any influence on the image quality, specifically preventing the image quality from degrading.
In order to achieve the foregoing objects, the present invention provides a method of judging digital watermark information which is embedded in at least one element of digital data by altering the at least one element of data.
The method comprises a first step of estimating a data value of the element by using data values of at least two neighboring elements of data located near the element; a second step of comparing the estimated data value with an actual data value of the element to detect the amount of the alteration thereof; and a third step of judging the presence and value of the digital watermark information based on the detected amount of the alteration. For example, the element data corresponds to pixel data, when the digital data is picture data, and the data value corresponds to a luminance value. The judgment carried out in the third step, for example, can be done by comparing a predetermined threshold with the amount of the alteration detected in the second step.
In recent years, the improvements in interpolation technology have been remarkable. When element data which constitutes digital data is lost, a lost element of data can be estimated from peripheral element data using such technology. In accordance with the present invention, data of an element, in which it is assumed that information is embedded, is estimated by using at least two elements of data near the element. The presumed data is used as reference data to compare to the actual data value of at least one element of data and to determine whether there is embedded digital watermark information and its value. Namely, it is judged whether or not information is embedded. The value is also determined when it is judged that information is embedded.
The above operation enables a more precise judgment than the conventional technology, since the presumed data value is nearly equal to the data value before embedding the information.